Development of audience design in adolescents' reference production

Page created by Dana Richardson
 
CONTINUE READING
Development of audience design in adolescents' reference production
Development of audience design in adolescents’
reference production
Caroline Arvidsson

Department of Linguistics
Bachelor’s Thesis 15 ECTS credits
Linguistics – Bachelor’s Course, LIN600
Bachelor’s Programme in Linguistics 180 ECTS credits
Spring semester 2021
Supervisor: Dr. Julia Uddén
Swedish title: Utveckling av mottagaranpassning i
ungdomars referentproduktion
Development of audience design in adolescents' reference production
Development of audience design in adolescents’
reference production
Caroline Arvidsson

Abstract
Compared to adults, children are less effective at designing their utterances to suit the informa­
tional needs of their audience. This listener­catering behaviour, known as audience design, has
been hypothesized to rely on domain general cognitive mechanisms, such as working memory
and cognitive flexibility. Considering that adolescence is an important period of sociocognit­
ive growth, research on the development of audience design beyond childhood is surprisingly
scant. The aim of this study was to trace the development of audience design in early and middle
adolescence, and test its reliance on cognitive control function. Participants (11–12 and 15–16
years) performed two tasks assessing (1) the ability to adjust referential expressions to inferred
knowledge of hearers and (2) cognitive control function. The findings suggest that the ability
to take into account the informational needs of listeners during utterance formation develops
considerably between early and middle adolescence. Although performance on both tasks was
higher in the middle adolescent group, the study provides no evidence for a reliance of the
measured audience design behaviour on cognitive control function. Future research should aim
to determine whether the development of audience design in adolescence is facilitated by an
increased efficacy of knowledge state attribution processes.

Keywords
Audience design, pragmatic development, referential communication, adolescence.
Development of audience design in adolescents' reference production
Utveckling av mottagaranpassning i ungdomars
referentproduktion
Caroline Arvidsson

Sammanfattning
Förmågan att anpassa sina yttranden efter samtalspartnerns behov är mindre utvecklad hos barn
än hos vuxna. Beteendet att ackommodera lyssnaren vid yttrandeformulering benämns ofta
som mottargaranpassning. Mottagaranpassning i konversation har föreslagits vara avhängig
exekutiva funktioner, såsom arbetsminne och kognitiv flexibilitet. Med tanke på att ungdom­
såren är en viktig period för social och kognitiv mognad har anmärkningsvärt lite forskning
genomförts på utvecklingen av mottagaranpassning under ungdomsåren. Målet med studien
var att undersöka utvecklingen av mottagaranpassning i ungdomsåren och testa dess eventuella
avhängighet av exekutiva funktioner. Deltagare (11–12 och 15–16 år) genomförde två tester som
mätte (1) förmågan att anpassa referentiella yttranden till lyssnares förmodade omvärldskunskap
och (2) exekutiva funktioner. Resultaten indikerar att förmågan att anpassa sina yttranden efter
lyssnares förmodade omvärldskunskap utvecklas betydligt under ungdomsåren. Trots att den
äldre åldersgruppen presterade bättre på testet som mätte exekutiva funktioner, predicerade inte
exekutiva funktioner förmågan att mottagaranpassa referentiella yttranden. Framtida studier
bör undersöka huruvida förmågan att tillskriva kunskapstillstånd till andra effektiviseras under
ungdomsåren, och således främjar utvecklingen av mottagaranpassning.

Nyckelord
Mottagaranpassning, pragmatisk utveckling, referentiell kommunikation, ungdomar.
Contents
1   Introduction                                                                                                                  1

2   Background                                                                                                                    2
    2.1 Audience design in reference production . . . . . . . . . . . . . . . . . . . . .                                         2
        2.1.1 A novel paradigm for modelling common ground on the inferred frame
                of reference of hearers . . . . . . . . . . . . . . . . . . . . . . . . . .                                       3
    2.2 The development of audience design in referential communication . . . . . . .                                             4
        2.2.1 Theory of Mind in childhood and adolescence . . . . . . . . . . . . . .                                             5
        2.2.2 Usage of Theory of Mind during online referential communication . . .                                               6
    2.3 The role of cognitive control function in online audience design . . . . . . . .                                          7
        2.3.1 A trade­off between formulating listener­appropriate expressions and
                speaker ease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                        8
    2.4 Aim and research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                        9
        2.4.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                         9
        2.4.2 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                          9

3   Method                                                                                                                       10
    3.1 Participants . . . . . . . . . . . . . . . . . . .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   10
    3.2 Materials and procedure . . . . . . . . . . . .      .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   10
    3.3 Tasks . . . . . . . . . . . . . . . . . . . . . .    .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   10
        3.3.1 Test of audience design (AD) . . . . .         .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   10
        3.3.2 Test of cognitive control function (CCF)       .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   12
    3.4 Analysis . . . . . . . . . . . . . . . . . . . . .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   .   13

4   Results                                                                                                                      14
    4.1 Preanalysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                      14
    4.2 Audience design as a function of age . . . . . . . . . . . . . . . . . . . . . . .                                       14
    4.3 Reliance of audience design on cognitive control function . . . . . . . . . . . .                                        15

5   Discussion                                                                                                                   16
    5.1 Possible limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                     17
    5.2 Ethical considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                     18
    5.3 Further inquiries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                    18

6   Conclusions                                                                                                                  20

References                                                                                                                       21

Appendix A Participant forms – Consent and project description                                                                   26
  A.1 Parents of participants under 15 years . . . . . . . . . . . . . . . . . . . . . .                                         26
  A.2 Participants under 15 years . . . . . . . . . . . . . . . . . . . . . . . . . . . .                                        28
  A.3 Participants 15 years and older . . . . . . . . . . . . . . . . . . . . . . . . . .                                        29
1        Introduction
When participating in conversation, speakers are continuously faced with the task of adapting
their utterances in accordance with the specific needs of interlocutors. This communicative
perspective­taking during online 1 language production is termed audience design (AD) (Clark
and Murphy, 1982), alternatively recipient design (Sacks and Schegloff, 1979). Models of AD
involve listener­catered adjustments that operate on all levels of language (Ferreira, 2019). This
can, for instance, translate to style­shifting (Bell, 1984), the adding and omitting of confusion
reducing content (such as case markings) (Kurumada and Jaeger, 2015), or the modification of
acoustic features in the speech signal (e.g. reduction vs. prominence) (Arnold et al., 2012).
One of the most salient listener­catering behaviours is the tendency to be particularly inform­
ative towards audiences that are assessed to need additional information to understand, such as
novices (Isaacs and Clark, 1987) or children (Tippenhauer et al., 2020). In reference produc­
tion, speakers who provide the information necessary to prevent miscommunication adjust their
referential expressions in accordance with the knowledge state of their listener. For example,
a speaker addressing an audience that is assumed to have no knowledge of Swedish politicians
may choose to utter ‘The prime minister of Sweden’, over ‘Stefan Löfven’, in order to increase
the possibility of being understood by that particular audience.
    Although children as young as 3–4 years old are aware that others’ knowledge states can
differ from their own (Perner and Leekam, 1986), 5– to 6­year­olds fail at using information
about listeners’ knowledge states in referential production tasks (Nadig and Sedivy, 2002). In
consequence, it has been suggested that the ability to use information regarding one’s listener’s
knowledge during online communication demands cognitive effort, and thus relies on domain
general cognitive control functions (CCF) 2 (Nilsen and Fecica, 2011). However, the validity of
traditional paradigms for measuring AD have been called into question, since optimal perform­
ance on these tasks can be reached by using visual selective attention alone (Rubio­Fernández,
2017). Considering that the ability to engage in communicative perspective­taking affects the
well­being of adolescents (Nilsen and Bacso, 2017), research on the development of AD in ad­
olescence is surprisingly scarce. This essay aims to address the gap in the literature by tracing the
development of AD in reference production and investigate its reliance on CCF in early and mid
adolescence, using an innovative design (Bentdz et al., 2020) that requires speakers to consider
the knowledge states of listeners to produce perspective­appropriate referential expressions.

    1
        The term ‘online’ denotes the actual time under which a given process takes place.
    2
        CCF is here used as a term interchangeable with executive function.

                                                           1
2        Background
2.1 Audience design in reference production
In every day dialogue, speakers often tailor their utterances to suit the informational needs of
their audience. This listener­catering behaviour is termed Audience Design (AD). Consider the
following exchange by two teenagers, overheard at an underground station in central Stockholm:

A:       that guy added me on snapchat
B:       who?
A:       that guy we met at Gabriel’s place
B:       aha
B:       I think his name is Johan

In this exchange, A makes two attempts at referring to Johan. The first attempt is insufficient
since it does not establish Johan’s identity, whereas the second attempt does. To most individu­
als, the referential expression in the second attempt is as ambiguous as the first; only a restricted
number of people are aware under what circumstances the two teenagers met Johan. B, however,
shares the knowledge needed to successfully interpret her conversational partner’s reference –
A’s second attempt is adequate because B is the addressee. Because of the simplicity and pervas­
iveness of referential communication (Asher, 1979), reference has received substantial attention
within AD research. This thesis is concerned with AD in reference production.
     Speakers that employ AD adhere to Grice’s (1975) Cooperative Principle. Namely, they
expect their utterances to be adequately informative, relevant and explicit in its particular con­
text. In regards to reference production, the Maxim of Quantity is perhaps the most applicable
one (p.45):

     1. Make your contribution as informative as required.

     2. Do not make your contribution more informative than is required.

The Cooperative Principle and the Maxim of Quantity state that speakers are expected to adjust
the informativeness of their utterances in commensurate with the informational needs of the
hearer. This includes not being overly informative. As an illustration, consider a situation in
which a speaker produces the referential expression ‘my oldest child’. If the speaker is talking
to a close family member, her referential expression is probably too informative. Conversely, if
she is talking to a stranger, ‘Erik’ provides too little information. AD entails providing the ideal
amount of information in a given communicative context. Clark et al. (1983) labels the optimal
informativeness of referential expressions the Principle of optimal design.
    AD requires the speaker’s ability to distinguish between what is common ground (knowledge
shared by the interlocutor) (Clark and Murphy, 1982), and what is privileged ground (know­
ledge exclusively available to the speaker) (Horton and Keysar, 1996). According to Clark and
Murphy (1982), common ground is accessed through a set of heuristics, including linguistic co­
presence (whether the referent has previously been mentioned in the discourse context), physical
co­presence (whether the referent can be perceived by both interlocutors), and group member­
ship 3 (whether a speaker and an addressee share knowledge or frame of reference by belonging
     3
     Clark and Murphy (1982) label this community membership, which can be considered as a somewhat narrow
term. Group membership will be used here, since it to a greater extent includes more general psychological features
(e.g. age) that are independent of which specific community a listener belongs to.

                                                        2
to the same community or group). In essence, by accessing the common ground through group
membership, the speaker does not rely on information regarding the physical or linguistic con­
text of the conversation, but rather on information about the inferred world knowledge or frame
of reference of their interlocutor. This information can be obtained by considering factors such
as the age, gender or cultural background of the listener.
    Taking the common and privileged ground into consideration during online communication
is a key aspect of successful reference production (Gillis and Nilsen, 2014). Primarily, stud­
ies on AD in reference have examined utterance modification with respect to speaker­listener
differences in physical co­presence and linguistic co­presence (Davies et al., 2016; Fukumura,
2016; Grigoroglou and Papafragou, 2019; Horton and Spieler, 2007; Loy et al., 2020; Nadig and
Sedivy, 2002; Nadig et al., 2009; Nilsen and Bacso, 2017; Nilsen and Graham, 2009; Saryazdi
and Chambers, 2020; Serratrice and De Cat, 2018). However, tasks designed to measure AD
by altering the physical (e.g. visual) co­presence of interlocutors have been criticized, since op­
timal performance in tasks of this kind can be achieved using selective attention alone (Rubio­
Fernández, 2017). An example of such a task is the well­established and commonly used Dir­
ectors task (Keysar et al., 2000). In this task, the speaker directs the addressee to move objects
along a grid of squares. The key feature of the Directors task is that some of the objects in the
grid are only visible to the participant (who can be either speaker or addressee). Participants who
base their interaction (e.g. utterance formulation or eye movement) on objects visible to their
confederate while disregarding objects in their own privileged view, have traditionally been un­
derstood as sufficient perspective­takers. Evidence at variance with this interpretation comes
from Rubio­Fernández (2017), suggesting that participants can employ a strategy of only focus­
ing on objects in white squares (visible to the confederate) and ignore objects in black squares
(not visible to the confederate), without actually reasoning about their partner’s perspective.
This brings in to question such tasks’ validity of measuring pragmatic ability. To address this
issue, Bentdz et al. (2020) created a task that required participants to take into account the world
knowledge (rather than visual perspective) of listeners in order to produce sufficient referential
expressions. This paradigm is presented in what follows.

2.1.1   A novel paradigm for modelling common ground on the inferred frame of reference
        of hearers
According to Clark and Murphy’s (1982) group membership heuristic, speakers take into ac­
count who their listener is during utterance formulation. A novel method of measuring this
behaviour was developed by Bentdz et al. (2020). The purpose was to create an AD task that
required participants to infer the knowledge state of their interlocutors. In their task, young
adults were instructed to verbally direct an addressee to choose a target item, which could either
be assumed to be known (known condition) or unknown (unknown condition) to the addressee
depending on their age, gender or cultural background. A successful method of denoting the
unknown targets was by describing them, in lieu of simply providing their names (e.g. ‘The ma­
chine with four propellers’ vs ‘The drone’, when directing a 91­year­old female hearer). Results
from Bentdz et al. (2020) indicate that adults take into account the identity of the hearer when
forming their referential expressions.
    Adopting a similar paradigm, Pagmar et al. (2021) found a different behaviour among 7­
year old children. Specifically, children chose one of the two strategies (describing vs naming,
although tentative results suggest naming was most prevalent in the unknown condition) and
continued to use that strategy throughout the task, regardless of condition. This suggests that
children do not adapt their utterances to the frame of reference of hearers. However, a major issue

                                                 3
interpreting the performance of the children in Pagmar et al. (2021) is that while their knowledge
of the targets was controlled for, their assumptions regarding the addressees’ knowledge of the
targets were not. Thus, it cannot be resolved whether the children maintained the same strategy
because (1) they consistently inferred that the listener had knowledge of the referent and hence
that describing the targets was redundant, or (2) they did not adapt their utterances in accordance
with their assumptions of the listener’s frame of reference. The former being an example of
failed inference regarding listeners’ knowledge states, whereas the latter exemplifies an absence
of AD­behaviour. These considerations underline the importance of controlling for participants’
assumptions regarding their listeners’ knowledge in similar paradigms.
    Studies investigating AD with respect to differences in interlocutors’ frame of reference/world
knowledge have either been conducted on children (Pagmar et al., 2021) or adults (Bentdz et al.,
2020). Little is known regarding how the ability to use information about others’ knowledge
states during online production develops in adolescence. The cross­sectional comparison of the
two age groups (11­12 and 15­16) in the current experiment moves beyond previous work, by
investigating the development of AD in early and middle adolescence.

In summary, AD is the tendency to tailor one’s utterance to suit the specific needs of one’s
interlocutor. It includes adjusting the informativeness of referential expressions in commen­
surate with listeners’ assumed frame of reference – a task that requires the speaker to consider
the listener’s level of access to the target referent and furthermore create a model for what is
common ground (knowledge shared by both interlocutors). If the speaker infers that the listener
does not share knowledge about a given referent, she might increase the informativeness of her
referential expression to provide the information necessary for successful denotation. Previous
studies suggest that factors such as listener age, gender or cultural identity generally affects
the informativeness of adults’ (Bentdz et al., 2020), but not children’s (Pagmar et al., 2021)
referential expressions. However, results from children (i.e. Pagmar et al., 2021) are difficult
to interpret since their assumptions regarding their interlocutors’ frame of reference were not
controlled for. Little is known with regards to how AD in reference develops throughout ad­
olescence. In the following sections, the literature on the development of perspective­taking in
referential communication is reviewed.

2.2 The development of audience design in referential communication
Piaget (1928/2002) claims that children under the age of 7 virtually ‘speak to themselves’ (p.
206). By that, he means that children are unable to adapt their utterances in accordance with
others’ knowledge states. Giving support to this notion, research has found that 5– to 6­year­old
children frequently fail to use information regarding their interlocutor’s perspective during ref­
erential communication tasks (Nadig and Sedivy, 2002). Despite this ‘egocentric’ behaviour,
children as young as 3 years old reason about others’ mental states and recognize that these
mental states can differ from their own (Perner and Leekam, 1986). This ability is known as
mentalizing or Theory of Mind (ToM) (Astington et al., 1988). Albeit typically developed chil­
dren over 7 pass most existing ToM tests (Peterson and Wellman, 2019), work on the social
cognitive development beyond childhood shows that adolescence represents a period of sig­
nificant growth in perspective­taking efficacy (Choudhury et al., 2006), and that activation in
brain regions involved in assessing common ground (Vanlangendonck et al., 2018) change in
adolescence (Blakemore, 2008). The main goal of this section is first to give a brief overview
of the development of ToM in childhood and adolescence, and second to account for previous
research on the development of perspective­taking during online referential communication.

                                                4
While views regarding at which age adolescence is reached vary across cultures, most re­
searchers agree that its commencement coincide with the onset of puberty (Blakemore, 2008).
The offset of adolescence is less clear. Although there is no widely accepted chronological
definition of the sub­stages of adolescence (Curtis, 2015), three phases of transitional youth will
be used in this thesis for the purpose of clarity: early adolescence ranges from 11 to 13 years,
middle adolescence from 14 to 17 years, and late adolescence/early adulthood reaches from 18
to 24 years (Curtis, 2015).

2.2.1   Theory of Mind in childhood and adolescence
In order to consistently produce perspective­appropriate referential expressions, speakers must
be able to assess the informational needs of their interlocutors. This entails recognizing that
others’ knowledge states can differ from their own. Children exhibit this latter ability from a
very young age (Wellman, 1992). For instance, Perner and Leekam (1986) showed that 3­ to
4­year­olds were able to keep track of and reproduce specific information that a play­partner
had missed due to being out of the room.
    ToM is a collection of concepts that allow individuals to perceive others as mental beings
driven by their own emotions, perspectives, attentional states, beliefs and desires (Astington
et al., 1988, p.3). Despite the lack of consensus regarding the basic characteristics of ToM
(Wellman, 1992), many researchers agree that concepts found to comprise the construction are
acquired sequentially and consistently in children around the world (Callaghan et al., 2005;
Shahaeian et al., 2011; Wellman and Liu, 2004). First, children learn that others have individual
beliefs, desires and access to knowledge. Second, they recognize that others can have false
beliefs and are able to conceal emotions. In typically developed children, these five concepts are
achieved before the age of 6 (Wellman, 1992). However, longitudinal behavioural data reveals
that ToM skills continue to develop steadily throughout childhood. For example, in addition to
providing support for the sequential acquisition of the five above mentioned concepts, Peterson
and Wellman (2019) demonstrated that the ability to understand sarcasm (i.e. that others can
mean the opposite of what they are saying) continues to develop in typically developed children
between the ages 10 and 11. This suggests that more intricate ToM­processes develop well after
the pre­school years, during a time when school children’s social world becomes increasingly
peer­oriented and complex.
    With respect to mentalizing abilities in adolescence, evidence from neuroscience has shown
that brain regions key to ToM processes (such as the medial prefrontal cortex and temporo­
parietal regions) undergo structural and functional changes during adolescence (Blakemore,
2008). This is especially noteworthy in the scope of the current study, since activation in the
medial prefrontal cortex was observed during a referential communication task that involved
accessing common ground information (Vanlangendonck et al., 2018). Furthermore, Bentdz
et al. (2020) found activation in these areas during online pragmatic comprehension. But how
do these changes in the activity and structure of the neural substrates involved in mentalizing
processes affect behaviour? Since children entering school age perform well on most ToM tasks
(Peterson and Wellman, 2019), designing experimental paradigms that do not generate ceiling
effects has been a challenge for behavioural scientist trying to trace the development of ToM
beyond childhood (Blakemore, 2008). However, data suggest that individuals become increas­
ingly effective at taking others’ (emotional) perspective throughout adolescence, as reflected by
shorter reaction times and less variation within age groups (Choudhury et al., 2006). Much like
the literature regarding the development of ToM in adolescence, the literature on the develop­
ment of adolescents’ online ToM­usage in communicative contexts is surprisingly scarce. In the

                                                5
following section, research on the development of ToM­usage in interaction is reviewed.

2.2.2   Usage of Theory of Mind during online referential communication
Compared to adults, children are less effective when it comes to making their referential expres­
sions as informative as required. In a study conducted by Nadig and Sedivy (2002), 5–6­year
old children and adults participated in a task designed to test speakers’ sensitivity to the visual
perspective of their interlocutors. In this task, participants instructed an experimenter to re­
trieve a target item from a display case. In the common ground condition, modifying adjectives
were necessary in order to successfully disambiguate the target from other competing items on
display. The results showed that while adults provided unambiguous utterances in 100% of tri­
als, children produced unambiguous descriptions in 75% of them. In a similar study involving
4–5­year old children, only 39% of the referential expressions were unambiguous (Nilsen and
Graham, 2009). These studies suggest that AD in reference production develops successively
during childhood, and that children reaching school­age have not yet developed into adult­level
proficient referential communicators.
     While there is considerable evidence that children are unable to fulfill the informational
needs of listeners during reference production, less is known regarding how AD in reference
develops throughout adolescence. In one study conducted by Fukumura (2016), 11­16 year old
participants (Mean: 13) were asked to describe a target picture that was presented together with a
competitor picture. The competitor picture only differed from the target in size. In the common
ground condition, the size­competitor was visually available to the addressee, and thus the use of
disambiguating size­adjectives was necessary to adequately denote the target. In the privileged
ground condition, the use of size­adjectives was redundant from the addressee’s perspective. The
results revealed that adolescents provided disambiguating adjectives nearly as often as adults
when necessary, but failed to omit size­adjectives when they were perspective­inappropriate.
Although Fukumura (2016) did not investigate the development of AD in adolescence (i.e. how
the performance of 11­year­old participants differed from the 16­year old participants’), the
results nevertheless indicate that the ability to take into account information concerning the
listener’s perspective during online production continues to develop in adolescence.
     Additional evidence that the use of ToM during online communication develops throughout
adolescence comes from comprehension tasks. For example, Dumontheil et al. (2010) demon­
strated that early adolescents (11­13 years) did not perform as well as middle adolescents (14­17
years) when interpreting referential expressions that required participants to take the visual per­
spective of interlocutors into consideration. Moreover, adults’ performance was better than the
performance of the middle adolescent group, which suggests that the online usage of ToM con­
tinues to develop in late adolescence. Whether this successive developmental pattern of com­
municative perspective­taking in comprehension corresponds to the development in production
can still be considered an unresolved issue.
     Speculating an informative model for adolescents’ AD progress, a fundamental question
arises: does the development of AD in reference production exhibit a steady change during
adolescence? More specifically, how does AD in reference production compare between early
and middle adolescents? Conceivably, the ability to use ToM during online production mirrors
the development in comprehension (as reported in Dumontheil et al., 2010), by successively
increasing throughout adolescence. Alternatively, advances may taper off in early adolescence,
suggesting that online ToM­usage demands distinct processes in production and comprehension.
The current study is the first to address this question, using an experimental paradigm that require
speakers to take into account the world knowledge, rather than the visual perspective, of hearers

                                                 6
in order to produce sufficient referential expressions.

This section has been concerned with the development of ToM and its usage during online ref­
erential communication. When children become of school age, they perform well on most ToM
tasks (Peterson and Wellman, 2019), but have not yet fully mastered the ability to use common
ground information in order to make their referential expressions as informative as required.
This inability becomes apparent in children’s tendency to generate ambiguous statements in
referential communication tasks (Nadig and Sedivy, 2002). Albeit the literature on AD in the
reference production of adolescents is scarce, a composite of results from production and com­
prehension tasks suggest that the ability to use ToM during online communication develops
successively throughout childhood and adolescence, into early adulthood.
     Because children show awareness of others’ perspectives (Perner and Leekam, 1986), but
fail to use this information during conversation (Nadig and Sedivy, 2002), it has been hypothes­
ized that the online usage of ToM relies on domain general cognitive mechanisms (Nilsen and
Fecica, 2011). In the next section, this hypothesis is considered.

2.3 The role of cognitive control function in online audience design
The term cognitive control function (CCF) 4 denotes a set of domain general, higher order pro­
cesses that facilitate flexible and goal­oriented behaviour (Miller and Cohen, 2001). In novel
or demanding situations, CCF is imperative to successfully controlling behaviour and thought
(Stuss, 1992). The increased efficiency of CCF throughout childhood and adolescence has been
interpreted as a manifestation of cognitive and emotional development (Diamond, 2002).
    Among the literature on the links between CCF and pragmatic development, research on
working memory, inhibition and cognitive flexibility is most prevalent (Matthews et al., 2018).
Working memory allows individuals to maintain and update information online (Diamond, 2002),
inhibition authorizes the restraint of behavioural and mental impulses (Leon­Carrion et al.,
2004), and cognitive flexibility enables the shifting of attention and selection of information
(Deák, 2004). In support of the rather intuitive notion that these processes assist in guiding up­
coming responses in communicative contexts, a meta­analytic study found a positive correlation
between measures of CCF and global measures of pragmatic ability in children (Matthews et al.,
2018). The study however concluded that that the relation between CCF and specific pragmatic
abilities is still unsettled.
    A core question in the referential communication literature is why children, and to some
extent adolescents, are incapable of effectively adjusting their referential expressions to meet
the needs of listeners. This inability is particularly puzzling given that children as young as 3–4
years old are able to understand that others’ knowledge states can differ from their own (Perner
and Leekam, 1986). One posited theory is that usage of ToM in conversation demands cognitive
effort, and that domain general cognitive mechanisms play a crucial role in facilitating mental
state attribution during online communication (Nilsen and Fecica, 2011). Because CCF is not yet
fully developed in adolescence (Anderson et al., 2001), taking into account the knowledge state
of listeners during utterance formulation may require too much cognitive load on individuals
without sufficient CCF. In what follows, a theory of a trade­off between AD­behaviour and
speaker ease is presented.
   4
       Again, note that CCF is here used interchangeably with executive function.

                                                         7
2.3.1   A trade­off between formulating listener­appropriate expressions and speaker ease
Communicative perspective­taking may seem effortless. Conversely, it is a complex process in
which interlocutors are required to manage the flow of linguistic, social and contextual input,
while conjointly planning their conversational contributions (Nilsen and Fecica, 2011). An ex­
ample of speakers’ tendency to prioritize ease over AD during utterance formation comes from
Brennan and Clark (1996). In the initial trial of their referential communication task, speak­
ers needed to employ a certain level of specificity, since lexemes denoting competing referents
were co­hyponyms (e.g. ‘birch’, ‘hazel’ and ‘willow’ are co­hyponyms of the hypernym ‘tree’).
In subsequent trials, labels of the competing objects did not share semantic fields (e.g. ‘tree’,
‘shoe’ and ‘car’), and thus, the same level of specificity was redundant. Despite this redundancy,
speakers adopting the appropriate level of specificity in the initial trial continued to be specific
when more general expressions would have sufficed. This indicates that the eventual pay­off
of switching strategies (adjusting the informativeness of referential expressions in adherence to
Clark et al.’s (1983) Principle of optimal design), was not worth its processing costs.
     During recent decades, a model concerning the trade­off between salience in meaning and
speakers’ processing efforts in communication has emerged (see, for instance Bacso and Nilsen,
2017; Gillis and Nilsen, 2014; Nilsen and Bacso, 2017; Nilsen and Fecica, 2011; Nilsen et al.,
2015). According to this framework, in order to produce listener­appropriate referential expres­
sions, the speaker has to: (1) suppress an ‘egocentric’ perspective (inhibition) and (2) switch to a
listener­friendly perspective (cognitive flexibility), while (3) maintaining and updating inform­
ation relevant to model common ground (working memory) (Matthews et al., 2018). Pursuant
to this model, if the cognitive load of managing the information flow becomes too heavy, the
speaker might revert to the ‘egocentric’ behaviour of producing an utterance that does not suit
the informational needs of her interlocutor (Epley et al., 2004). Accounts from meta­analytic
studies report that children with ASD or ADHD, which are diagnoses typically associated with
diminished pragmatic skills (Loukusa et al., 2018; Staikova et al., 2013), exhibit deficits in CCF
(Demetriou et al., 2019; Willcutt et al., 2005). These deficits have furthermore been related to
why children with ADHD and ASD do not perform as well as their typically developed peers in
referential communication tasks (Dahlgren and Sandberg, 2008; Nilsen et al., 2013).
     Evidence from direct measures on the relationship between CCF and the ability to adjust the
informativeness of referential expressions is mixed. Nilsen et al. (2015) found a link between
measures of 9­ and 12­year­olds’ CCF (working memory and inhibition) and their tendency to
provide unambiguous utterances. However, in a similar study on younger children (years: 4­5),
no link between CCF (working memory, inhibition and cognitive flexibility) and the ability to
provide sufficiently detailed referential expressions was observed (Nilsen and Graham, 2009).
On the adult side, Horton and Spieler (2007) found that young adults (N=24) were better at ad­
justing the informativeness of utterances to suit their conversational partner than elders (N=24),
which, combined with group­level measures of working memory, led the authors to conclude
that the lack of AD in elders’ utterances may reflect the difficulty of accessing listener­specific
information from memory during online communication 5 . Contrary to this, Bentdz et al. (2020)
found no correlation between adults’ performance on a referential production task and their
working memory capacity. Note that Bentdz et al. (2020) had a considerably larger sample size
(N=201) than the study that reported correlations.
     In attempts to explain the mixed findings regarding the role of CCF in referential commu­
nication, researchers have proposed that this role varies across the life span (Matthews et al.,
   5
    It should be noted that the mean working memory scores for the younger adult population were retrieved from
another sample (i.e. not participants in the referential communication task).

                                                      8
2018; Nilsen and Graham, 2009). In other words, it is possible that some CCF subfunctions
play a central role in communicative perspective­taking during adolescence, but not early child­
hood. Because Nilsen et al. (2015) found a correlation between measures of 9­ to 12­year old
participants’ CCF and the ability to produce listener­appropriate referential expressions, such a
link may be expected among early adolescents in the current study. Whether this link remains in
middle adolescence, which is an important period of social and cognitive growth (Blakemore,
2008), has to the best of my knowledge not been tested.

To summarize, CCF is a set of processes vital to guiding thinking and behaviour (Miller and
Cohen, 2001). It has been suggested that CCF enables the usage of ToM information during
communication (Nilsen and Fecica, 2011). According to this model, speakers may revert to
an ‘egocentric’ behaviour in favour of processing ease, when the demands of managing the in­
formation flow in conversation become too heavy (Nilsen and Fecica, 2011). Evidence for a
relationship between the ability to produce listener­appropriate referential expressions and CCF
is mixed (Matthews et al., 2018), and little is known regarding this relationship in adolescence.
In pursuance of contributing to the understanding of online AD’s reliance on CCF, and address­
ing the issue of how this eventual reliance develops across the life span, the current study will
examine the link between AD behaviour in reference production and CCF in adolescence.

2.4 Aim and research questions
2.4.1   Aim
The aim of the present study is to trace the development of audience design in reference produc­
tion during early and middle adolescence (years: 11­12 and 15­16) and investigate its reliance
on cognitive control function (CCF).

2.4.2   Research questions
   1. Does the ability to adapt utterances to hearers’ assumed knowledge states increase as a
      function of age in early and middle adolescence?

   2. Do measures of adolescents’ CCF predict their ability to adapt utterances to hearers’ as­
      sumed knowledge states?

        (a) Do measures of CCF predict the ability to adapt utterances to hearers’ assumed
            knowledge states in early adolescence?
        (b) Do measures of CCF predict the ability to adapt utterances to hearers’ assumed
            knowledge states in middle adolescence?

                                               9
3       Method
3.1 Participants
The final sample consisted of 59 adolescents from two age groups (11­12 years: N = 30, 18 girls;
15­16 years: N = 29, 17 girls), recruited from two schools in Värmdö municipality, Stockholm.
By disseminating information via the participating schools, parents of children in the younger
age group who were willing to allow their child to participate were first recruited. Participants
in the older age group were recruited directly via their school. To ensure that participants had
Swedish as one of their dominant languages, being a fluent speaker of Swedish at 7 years old was
a criteria 6 . Five additional adolescents participated, but were excluded from the analysis due
to having been diagnosed with ASD, ADHD and/or DLD (Developmental Language Disorder).
Informed consent was obtained from the legal guardian(s) of participants in the younger age
group, as well as all participants. Upon participating, the adolescents in the older age group
received a gift card to a value of 50 SEK, and the participants in the younger age group were
offered confections. The study was approved by the Swedish Ethical Review Authority (Dnr
2020­07083). Information and consent forms provided to participants are available in Appendix
A.

3.2 Materials and procedure
During one full half hour session, the participant was presented with two tasks respectively de­
signed to measure (1) audience design (AD) in reference production and (2) cognitive control
function (CCF). The AD task was always the first administered, since it was considered the most
important task in regards to the research questions. Throughout the experiment, the participant
was seated at a table beside an experimenter and in front of a laptop. In the AD task, the parti­
cipant was recorded using the laptop’s built­in microphone. The experiment was conducted in
a quiet room at the participant’s school.

3.3 Tasks
3.3.1    Test of audience design (AD)
The referential production task used in the present study was adopted from Bentdz et al. (2020)
and Pagmar et al. (2021). The aim of the task was to measure adolescents’ ability to adapt their
referential expressions to the inferred frame of reference of their addressee. Successful reference
entailed providing at least one visual feature of a referent, rather than for example using its name
(e.g. ‘The blue bird’ vs ‘The Twitter logo’), in cases where the addressee was not assumed to
have knowledge of the referent.
    In each trial, a picture of the addressee (either a child or an elderly woman) was presented in
the upper corner of a laptop screen. At the bottom of the screen, one target (e.g. a video game
character, a musician, or other individual/object) and three competitor pictures were displayed.
In both of the task’s conditions, the speaker knew the name of the target but assumed that the
addressee either (1) did not know the name of the target (unknown condition) or (2) did know
the name of the target (known condition). An example of four trials from the task’s two condi­
tions is given in Figure 1. Participants’ knowledge of each target referent and their assumptions
    6
    This criteria was considered a straightforward approach towards ensuring that all participants had equal op­
portunity to understand the information provided during the course of participation.

                                                      10
Figure 1: Examples of four trials showing the AD task’s two conditions with two addressees. In
the exemplified known trials, the targets are a Christmas tree and a Santa Claus. The targets in
the unknown conditions are a Minecraft character and the Twitter logo.

regarding the listener’s knowledge of each target referent were controlled for through post­test
surveys.
     During the practice trial, the participant was instructed to say something that would enable
their addressee to choose the target picture. The practice trial was always a known condition
trial (where both the speaker and addressee knew the name of the referent and thus referring
by describing was unnecessary). This choice was based on pilot runs of the experiment, in
which only participants that were presented with an initial unknown condition trial described
referents throughout the entire experiment, irrespective of trial condition 7 . Interestingly, the
AD task from Pagmar et al. (2021) (where a portion of the participating children described
referents throughout) had a practice trial that required them to describe the referent in order
to succeed. If a participant during the current experiment’s practice trial chose to describe the
target (again, describing was redundant in the practice trial), for instance, ‘the tree with a star’,
the experimenter always uttered: ‘Yes, and you could also just say Christmas tree, right?’. This
was deemed necessary in order to avoid the risk of having any participant misunderstand the
task, thinking that they were expected to consistently describe the target.
     Unlike Bentdz et al. (2020) and Pagmar et al. (2021), stimuli in the current task were se­
lected to suit the frame of reference of adolescents. In as many unknown condition trials as
possible, it was vital that the participants would know the name of targets, as well as assume
that the listeners did not know the target names. To achieve this, six 11­year­old individuals
(that were not part of the experiment) were recruited through their parents to complete an online
survey. The parents first received information about the survey through social media posts. The
children were presented with pictures of potential targets and asked to (1) provide the names
   7
     The tendency to stick to a single strategy (i.e., either only describe or only use the target name) was not observed
in sessions where the initial trial was a known condition trial.

                                                          11
Table 1: Translated examples (from Swedish) of descriptions of target referents. To provide
referential expressions that successfully distinguished targets from competitors, a noun phrase
(NP) with a maximum of one modifier (an adjectival phrase (AP) or a prepositional phrase (PP)
was necessary.
 Target                        Competitors                                 Example of target description
 Minecraft character 8         Cartoon figures (none in t­shirt)           ‘The man in a t­shirt’
 Twitter logo                  Pink, white or greenish birds               ‘The blue bird’
 Christmas tree                Trees                                       ‘The tree with a star’
 Santa Claus                   Elderly males without beards                ‘The man with a beard’

of the referents (by typing the names in a text box) and (2) answer whether they thought that
the elderly and child addressees knew the names of the referents. This allowed for deciding
which targets might have a higher chance of being known to adolescents in general, as well
as which targets were suitable in the unknown condition trials. Furthermore, in an attempt to
control for processing efforts of describing targets, the competitor stimuli were chosen so that
an NP containing no more than one modifier (e.g. an AP or a PP) was enough to comprise an
unambiguous target description. An example of this is given in Table 1. Altogether, the task
consisted of 36 trials (24 unknown condition, 12 known condition). Four randomized versions
of the experiment were presented in an evenly balanced fashion across participants. The test
was implemented using PsychoPy (Peirce, 2007). Since participants varied in number of test
and control trials (depending on what they had answered in the post­test survey), the AD­score
for each participant was computed as follows:
                                                          Udescr Kdescr
                                          ADscore =             −                                            (1)
                                                           Utot   Ktot
where Udescr is the number of successful referential expressions (i.e. descriptions of targets
assumed to be unknown to the listener), Utot is the total number of targets that the participant
assumed were unknown to the listener, Kdescr is the number of redundant descriptions (i.e. de­
scriptions of targets assumed to be known to the listener) and Ktot is the total number of targets
assumed to be known to the listener – resulting in an AD­score somewhere between ­1.0 and
1.0. The rationale for subtracting the ratios was to account for the possibility that the participant
used describing as an overall strategy (Bentdz et al., 2020).

3.3.2      Test of cognitive control function (CCF)
To assess adolescents’ CCF, the Wisconsin card sorting test (WCST) was administered. The
WCST is one of the most frequently used neuropsychological measures of CCF (Cummings,
2017). However, the relation between AD and CCF has to the best of my knowledge not previ­
ously been analyzed by usage of the WCST. In this task, participants matched cards according to
a hidden matching rule (colour, number or shape) that changed after 10 consecutive trials. The
task required participants to figure out the rule by trial and error with feedback responses after
each matching. WCST performance generally correlate with CCF subfunctions frequently hy­
pothesized to predict AD in reference production (working memory, cognitive flexibility and in­
hibition) (Lehto, 1996; Steinmetz and Houssemand, 2011; Van Eylen et al., 2011). The perform­
ance of children with ADHD and ASD have been shown to exhibit lower WCST performance
   8
       The experimenter learned from the participants that the name of this character is actually ‘Steve’.

                                                          12
than that of typically developed children (Tsuchiya et al., 2005). Because children with ADHD
and ASD exhibit difficulties in referential communication tasks (Dahlgren and Sandberg, 2008;
Nilsen et al., 2013), WCST performance is of particular interest in the current study.
    Two WCST scores were utilized: % perseverative errors and % random errors. A persev­
erative error occurs when a card is incorrectly matched according to a previous rule although
a rule switch has been indicated by the feedback response. A random error (also known as a
non­perseverative error) is an error that is not perseverative or efficient (i.e. an error associated
with efficient hypothesis testing immediately after a rule change) (Barceló and Knight, 2002).
Conceptually, perseverative errors represent difficulty in flexibly switching matching behaviour
after a rule change, and random errors reflect difficulty in maintaining successful matching be­
haviour during a current rule (Barceló and Knight, 2002). Among the three CCF subfunctions
mentioned in this thesis (working memory, inhibition and cognitive flexibility), perseverative
errors and random errors have both been found to be best predicted by measures of cognitive
flexibility and working memory in adolescence (Huizinga et al., 2006; Huizinga and Van der
Molen, 2007). Percentage of a given error type was calculated by taking the number of errors of
that type divided by 60, which was the total number of experimental WCST trials. The compu­
terized implementation of WCST used in the current investigation was available at PsyToolkit’s
(Stoet, 2017; Stoet, 2010) experiment library 9 .

3.4 Analysis
Most statistical analyses were conducted using univariate generalized linear modelling (GLM).
The GLM is an extension of regression models that can be utilized to fit both quantitative and
qualitative predictor variables (Haase, 2011, p.24). To test AD as a function of age, AD­scores
were analysed with the factor AGE GROUP (11–12 vs. 15–16). In addition, analyses were
performed to investigate the relationship between AD and CCF. In these models, the WCST
scores (% perseverative errors and % random errors) were independent, continuous variables.
All frequentist analyses were conducted with the statistical software R (R Core Team, 2021)
and an alpha­level of α = 0.05. The model parameters were fit using the built­in functions
glm (regression with categorical factor and/or continuous predictors), and aov (ANOVA with
categorical factor).
    A Bayesian independent samples t­test with grouping variable AGE GROUP and dependent
variable AD­score was conducted in JASP (JASP Team, 2020). Bayesian hypothesis testing is
used to assess the relative probability for competing hypotheses (Wagenmakers, 2007). Bayes
Factor 10 (BF10) was used to indicate the strength of evidence from the data for H1 (that the
two populations’ AD­score means are not equal). As an illustration, BF10 = 5 would indicate
that there is 5 times more evidence for H1 than H0. Bayesian hypothesis testing is dependent
on the hypotheses’ prior probabilities (Wagenmakers, 2007). In this study, the default prior in
JASP was used, and is described by a probability distribution centred around zero with a width
parameter of 0.707.

   9
       https://www.psytoolkit.org/experiment­library/

                                                        13
4        Results
4.1 Preanalysis
Audience design (AD) data from trials with target referents that the post test survey revealed to
be unknown to specific participants were excluded from the analysis (133 out of totally 2118
trials). The participants (N = 2) from whom most data points were excluded declared that they
did not know six of the target referents. A composite of all participants’ mean AD­scores was
as follows: M = 0.61, std. error = 0.04. A Shapiro­Wilks normality test showed that the AD
data was positively skewed (data were normally distributed for the younger age group, whereas
66% of the older adolescents had a score of 0.7 or higher, with the maximum and minimum
possible scores being 1.0 and –1.0). Thus, before being fit to the linear regression models, data
were transformed using a rank­based inverse normal (RIN) transformation 10 , which compared
to a wide range of transformation methods have been shown to be most beneficial with respect
to statistical power on moderate sample sizes (N ≥ 20) (Bishara and Hittner, 2012).

4.2 Audience design as a function of age
A univariate generalized linear model with factor AGE GROUP revealed a significant effect on
AD performance (b = 0.84, std. error = 0.16, t(57) = 3.68, p < .001). Modelling the non­
transformed data generated a similar result 11 . AD performance by age group was: 11–12
years: M = 0.48, std. error = 0.17, 15–16 years: M = 0.74, std. error = 0.17 (Figure 2). The

Figure 2: Results from the analysis of age group and AD performance. The red line represents
the development of mean ability to adapt utterances in commensurate with the assumed frame
of reference of the hearer. The circles represent the performance of individual participants.
    10
    All visualisations of AD performance in this section show non­transformed AD­scores.
    11
    Results of AD as a function of age before RIN transformation: b = 0.26, std. error = 0.07, t(57) = 3.32, p =
.002).

                                                      14
Bayesian independent samples t­test found strong evidence for a difference in mean AD­score
(non­transformed) for the two AGE GROUP levels (BF10: 20.93). Compared to their younger
peers, 15­16­year­olds described referents they assumed to be unknown to hearers more often (in
93% vs. 82% of unknown condition trials), and described referents they assumed to be known
to hearers less often (in 17% vs. 33% of known condition trials).

4.3 Reliance of audience design on cognitive control function
A univariate generalized linear model with WCST errors as independent variables and both age
groups’ AD­scores (RIN transformed) as dependant variable did not generate significant effects
(intercept: b = 0.25, std. error = 0.37, t(56) = 0.68, p = .49, % perseverative errors: b = –2.01,
std. error = 2.6, t(56) = –0.75, p = .45, and % random errors: b = –0.47, std. error = 2.68, t(56)
= –0.18, p = .86).
    The relationship between measures on CCF and AD performance within the individual age
groups was analyzed with two univariate generalized linear models (one per age group). The
WCST error scores were independent variables, and the RIN transformed AD­score was the
dependent variable in both models. No significant effect of WCST errors on the younger age
group’s AD performance was observed (intercept: b = –0.22, std. error = 0.58, t(27) = –0.38, p
= .7, % perseverative errors: b = –0.57, std. error = 3.62, t(27) = 0.86, p = .88, and % random
errors: b = 2.83, std. error = 3.62, t(27) = –0.16, p = .4). No significant effect of WCST errors
on the older age group’s AD performance was found (intercept: b = –0.31, std. error = 0.59,
t(26) = –0.53, p = .59, % perseverative errors: b = 2.03, std. error = 3.93, t(26) = 0.52, p = .61,
and % random errors: b = 0.37, std. error = 6.29, t(26) = 0.06, p = .95).
    Two follow­up one­way ANOVA tests showed that the amount of WCST errors differed
significantly between the age groups. Mean % perseverative errors was lower in the older age
group (11­12 years: M = 0.18; std. error = 0.01, and 15­16 years: M = 0.14; std. error = 0.01;
F(1,57) = 13.23, p = 0.005) Mean % random errors was also lower in the older age group (11­12
years: M = 0.12; std. error = 0.01, and 15­16 years: M = 0.07; std. error = 0.01; F(1,57) = 8.65,
p < .001). AD performance and amount of WCST error types per participant are visualized in
Figure 3.

Figure 3: Panel figure showing the relation between AD performance and WCST error types.
Visualized in the panels are individual participants’ AD­score in relation to % perseverative
errors (panel A), and AD­score in relation to % random errors (panel B).

                                                15
You can also read